3D Beamforming for Capacity Boosting in LTE-Advanced System

3D Beamforming for Capacity Boosting in LTE-Advanced System Hyoungju Ji, Byungju Lee and Byonghyo Shim Seoul National University, Seoul, Korea Email: {hyoungjuji, bjlee}@islabsnuackr, bshim@snuackr Young-Han Nam, Youngwoo Kwak, Hoondong Noh and Choelkyu Shin Samsung Research America Dallas, Dallas, USA Samsung Electronics DMC R&D Center, Republic of Korea {younghann, ywkwak, hoondongnoh, ck3shin}@samsungcom Abstract LTE-Advanced has been deployed with 2 and 4 transmission antennas (Tx) while the specification supports up to 8Tx Due to deployment space, antenna dimension and complexity, operators have not been interested in the deployment of 8Tx s Recently, three dimensional (3D) beamforming using 2D active antenna array has attracted significant attention in the wireless industry By incorporating 2D active array into LTE-A s, the offers freedom in controlling radiation on elevation and horizontal dimension In addition, 2D array antenna increases the number of antennas without exceeding form-factor where the conventional antennas are deployed When the number of antennas increases in the form of 2D arrangement, spatial separation can be realized simultaneously in horizontal and elevation domain and vertical beam-steering can increase SINR of UEs in high floors In this paper, we study the operations and implementations for supporting 3D beamforming with 8Tx antennas In our schemes, by reusing the conventional CSI feedback framework, the can operate 2D active array without harming the backward compatibility Evaluation results show that 3D beamforming provides capacity boosting over the conventional 2D beamforming s while keeping same antenna structure Keywords beamforming; LTE; Multi-user transmission; I INTRODUCTION In recent years, the wireless industry has witnessed a drastic increase of wireless data traffic on a global scale [] In response to the increasing demand for data traffic, 3rd Generation Partnership Project (3GPP) has initiated new standardization effort to provide cutting-edge techniques with an aim to improve spectral efficiency and user experience Among many features studied in the 3GPP for long-termevolution advanced (LTE-A) s, multi-input multi-output (MIMO) has been recognized as one of key technology to enhance the spectral efficiency [2] In the first release (Rel8) 2Tx and 4Tx MIMO s with spatial multiplexing and transmit diversity schemes has been introduced In the Rel, MIMO enhancement for 8Tx and UEspecific beamforming has been adapted For the next step of MIMO enhancements, full dimension MIMO (FD-MIMO) has received much attention in recent years Currently, study items have been initiated to identify key features to support up to 64 antennas placed in a 2D array structure [3] By incorporating FD-MIMO into LTE s, it is expected that the throughput will be improved drastically over the conventional LTE s Notwithstanding the rosy prospect, it is in practice not easy to deploy 8Tx antennas for various reasons One of the key reason is the deployment space and antenna size At the first stage of LTE deployment, most of operations have focused on 2 or 4 antennas in the enb side In fact, when the cross-polarization antenna is used, the required deployment space in the transmitter tower for GSM, WCDMA, and LTE s is more or less similar Thus, the operators can simply exchange antennas from the legacy to the LTE In addition, 8Tx antenna s do not provide attractive beamforming gain in dense urban scenario while offering substantial cost for doubling backhaul capacity Further, by blocking high-rised buildings, propagation channel experiences wide angular spread which introduces high interuser interference From these observations, it is clear that it is not so desirable to upgrade from 2 and 4Tx to 8Tx Vertical beamforming, placing additional antenna in vertical direction, is one of traditional yet attractive solution for supporting MIMO transmission on the limited deployment space Such as single-user MIMO (SU-MIMO) s, it is possible to achieve full rank of the channel matrix when the uncorrelated channel condition between large spaced vertical antennas is satisfied [4], [5] In [4], it is observed that the rank of channel matrix increases with antenna separation in vertical domain To achieve similar rank utilization with λ spaced horizontal antennas, the vertical antenna spacing should be 3 Fig 2 and 4Tx passive antennas (top), 2D 8Tx active antenna realizing transceivers in vertical and horizontal (bottom) 978--4673-6782-/5/$3 25 IEEE 2344

to 4 times wider than horizontal spacing [5] Owing to the active antenna technology, it is not possible to utilize multiple antennas placed in a 2D antenna array panel to realize 2-by-4 and 4-by-2 configurations without increasing total number of antenna elements In contrast to the passive antennas placed in a linear array, the antennas are placed in 2D planer array and further they are comprised with active radiation components A conceptual diagram of an enhanced 8Tx active antenna enb is shown in Fig 2D 8Tx enb has two distinctive features over the legacy MIMO s Firstly, the transceiver units can be placed in 2D antenna array panel to enable 3D beamforming without harming the backward compatibility Thus, downlink MIMO can support up to 8 antennas for the co-scheduled UEs in both vertical and horizontal direction using the conventional CSI feedback framework Secondly, to verify the potential benefit of 2D 8Tx s, careful and comprehensive evaluation of 3D beamforming implementation schemes is needed 3D deployment scenarios should be investigated to deploy urban environments where UEs are located in different floors The remaining part of this paper is organized as follows In Section II, we discuss the transceiver architecture to support 2D 8Tx Section III describes two operation schemes to enhance 2D beamforming to 3D beamforming In Section IV, we discuss the evaluation models and scheduling methods along with the evaluation results Finally, conclusions are drawn in Section V II 2D TRANSCEIVER ARCHITECTURE The 2D active antenna can be organized in three major units: transceiver unit, radio distribution network unit, and the 2D antenna array unit [6] The has the transceiver unit array which can be considered as logical antenna ports In the transceiver unit array, transmit or receive beamforming can be applied and UE can benefit from it Radio distribution network unit delivers transceiver signal to antenna elements Transceivers can be configured with required amplitude and phase weights such that one or more users are multiplexed from the 2D antenna array In this work, we consider FDD-based SU/MU-MIMO as shown in Fig 2 We assume that the enb has N T (=N V N H ) transceiver which are placed in M T (=M V M H ) antenna elements Consider a with N T transmit antennas at enb, K co-scheduled UEs with precoding weight W, the received signal y for the kth UE can be derived as Transceiver units Fig 2 V Radio distribution unit Antenna array unit System Model for 2D 8Tx PDL PDL y k = N T K H kg k W k x k + H k G l W l x l +n k N T K l k Desired signal MU interference () where P DL, H k C N R M T, W k C N T r, s k, n k CN(,σk 2 I) denote downlink transmission power, channel matrix between the enb and kth UE,user-specific beamforming matrix with{ rank r, transmit signal for the kth downlink user with E x k 2} =and the additive complex Gaussian noise with zero mean and variance N, respectively The radio distribution weight G k C M T N T, representing the relationship between the transceiver to the antenna element, is expressed as V k V k G k =, V k = [ ] T v,v,, v MT /N T V k (2) v i = exp ( j 2πλ ) MT /N (i ) d v cosθ t (3) T where i =,, M T /N T d v and θ t is vertical element spacing and cell-specific electrical vertical tilting angle, respectively By setting G k to a diagonal matrix, we can avoid the scenario that more than one transceiver use same antenna elements at the same time In order to reduce implementation complexity, we assume that enb uses the same distribution weight to each UE The antenna elements can be placed with a spacing of λ to support spatial multiplexing as well as beamforming transmission Each transceiver is connected to M T /N T antenna elements maintaining the aperture size of vertical dimension Then, the effective antenna spacing for precoding can be a multiple of λ As shown in Fig, vertically λ and 2 λ spacing can be maintained with and 8Tx, respectively The received signal for the kth uplink user at the enb is given by + P UL r l G T l h l d l +n k (4) u k = P UL r k G T k h k d k Desired signal l k } {{ } MU interference where P UL, h k C M T, r k C N T, d k denote the uplink transmit power, channel vector between the kth UE and enb, uplink reception vector, transmit signal for the kth uplink user, respectively In the TDD, channel state information is available at the enb using the channel reciprocity between uplink and downlink However, such property does not hold for FDD and hence a proper scheme to feed back the channel of antennas is required Clearly, downlink channel measurement and feedback should be essential for enabling 3D beamforming III 3D BEAMFORMING OPERATION IN LTE-A SYSTEMS In this section, we derive two schemes to enable 3D beamforming with 2D transceiver under the conventional CSI feedback framework First scheme () is to use two 2345

separate CSI feedbacks instead of using single 8Tx CSI feedback Each CSI report can be configured to measure different antenna sets in same enb antennas Note that the scheme is to reconstruct the channel matrix by employing Kronecker product of two partial measurements (both vertical and horizontal) The second scheme () is to exploit the conventional 8Tx codebook to supprt 2D antenna array In this case, UE can directly use the conventional CSI feedback framework although 8Tx codebook does not perfectly fit to 2D antenna array When compared to the legacy 8Tx, 2D transceiver does not bring any additional computational complexity in the uplink reception A Downlink Transmission ) : In LTE s, pilots for 2, 4 or 8 antennas can be configured to feed back CSI from UEs (called CSI process) and UE can report up to 3 different CSI processes for CSI feedback to enb [7] In LTE-A, multiple CSI process has been designed to support multi-point transmission and reception When UE connects two transmission points (TP), one CSI process is configured for the first TP and the other for the second TP for joint reception Since the antenna structure is unavailable to UE in the LTE-A s, CSI processes can be used for 2D antenna measurement One way is to configure two CSI processes for each dimension of antenna structure; one for horizontal CSI measurement and the other for vertical CSI measurement To express 2D array antenna, we can represent the channel matrix H {k,j} for each receive antenna at kth UE as h, h,2 h,nh H {k,j} h 2, h2,2 h2,nh = (5) h NV, h NV,2 hnv,n H for j =,,N R Then, UE can be configured one set of reference signal to measure CSI from one column of H and the second set to measure CSI from one row of H to horizontal CSI Overall channel H can be reconstructed with Kronecker product of two reported PMIs of each CSI measurements Then, output SINR of the MMSE-IRC receiver of the ith layer with the sth precoder can be expressed [8] γ k ρ k,i = [ ( ) ] (6) (h k g k w k,s) h k g k w k,s + γ k I ii where i r, s is precoder index of set S, γ k is SINR of kth UE and [] ii denotes the ith diagonal element Optimum rank r and precoder s can be selected to maximize the data rate as r {r k,s k } = arg max C (ρ k,i,s ) (7) r,s s S,i= where C(ρ k,i,s )= 2 log( + ρ k,i,s) In each subband, CQI can be selected from the effective output SINR of receiver as ( rk ) ρ subband,k = C 2 log ( + ρ k,i,s k ) (8) i= Since the UE does not have any knowledge of actual channel dimension of the, UE would separately report the preferred CSIs (RI, PMI and CQI) from two CSI processes Then, actual RI can be the maximum between vertical rank and horizontal rank For 2D PMI, the enb can generate 2D precoder using Kronecker product of two PMIs For 2D precoder of rank and 2 (we denot these as W (),2D, W (2),2D, respectively), Kronecker based codebook set can generates the combination of vertical and horizontal D PMI reports as W m,n (),2D W m,n (2),2D = W (),D = W (),D W (),D m h,n h, W (2),D m h,n h or W (2),D W (),D m h,n h (9) Composite subband CQI can also be calibrated with the product of two effective SINRs as ρ eff subband,k ρ(vertical) subband,k ρ (horizontal) subband,k However, the effective CQI would count intercell interference twice from each CSI process, there would be CSI mismatch between calibrated and actual CQI Further, UE would select best PMIs for each channel dimension and those may not the same as jointly selected PMI from full antenna measurement This degrades the link capacity when MU precoding is applied 2) : In LTE s, 8Tx codebook has been developed to adapt channels with D linear antennas For example, rank precoder can be expressed W m,n () = / 8[v m ϕ n v m ] T with v m =[ e j2πm/32 e j4πm/32 e j6πm/32 ] T and ϕ n =e jπn/2 and v m is used to select the beam direction with same polarized antennas and ϕ n can represent co-phase between different polarization antenna sets Even though 2D 8Tx has the same length of precoding weight, 8Tx codebook may not be feasible to 2D precoding weight Nevertheless, reusing the conventional feedback framework, 8Tx codebook can be re-mapped to 2D transceiver of and as (4V,2H) wm,n N V N H =/ 8[ (2V,4H) m,n N V N H v m +45 pol ϕ n v }{{ m ] } 45 pol w = [ e j4πm/32 e j6πm/32 ϕ n e j4πm/32 ϕ n e j6πm/32 ] 8 e j2πm/32 ϕ n ϕe j2πm/32 +45 pol 45 pol () For, 8Tx codebook weight can be transposed in vertical direction The transposed codebook can jointly but sparsely quantize vertical and horizontal angles Co-phase information can be reused between different polarization In case of, effectively λ and 2λ spacing is supported between horizontal transceivers and vertical transceivers, respectively Since the effective vertical spacing is four times larger than horizontal spacing, 8Tx codebook weight can be re-mapped to keep large phase shift π/8 between vertical transceiver and small phase shift π/6 between horizontal transceivers With scheme 2, precoder would not be optimized for 2D antenna array, but other CSI information can be reused without any changes Fig 3 shows the elevation beams where the proposed schemes are applied in 2D 8Tx array antennas B Uplink Reception For uplink transmission, enb can estimate the channel information at all received antennas using sounding signal n 2346

In the conventional 2Tx or 4Tx enb, therefore, each antenna only takes combining gain with a fixed beam in the vertical domain In the cases of 2D transceiver configurations, on the other hand, there is a tradeoff between vertical combining gain and dynamic beam control in vertical domain depending on how each column is divided by transceiver For instance, although the antenna configurations in Fig have the same number of 2D antenna element, they would have different vertical combining and dynamic beam control with combining horizontal domain IV EVALUATIONS A SU/MU Adaptive Scheduling In each scheme, SU and MU adaptive scheduling is applied for each subband in both downlink and uplink In case of downlink, rank adaptation is considered for SU-MIMO and maximum two UEs can be multiplexed with rank- for MU- MIMO In case of uplink, only rank is assumed for both SU and MU-MIMO scheduling For SU/MU-MIMO adaptation, UE set k i,k j are optimized per subband as {k i,k j } K K { ( ( = argmax C ρ eff subband,k i )+ C ρ eff subband,k j k )} k i= k j=,k i k j () B 3D Deployment and Channel Model In urban deployments, most of the UEs are located indoors on different floors, Considering this, two main scenarios are currently under discussion [9] The first scenario is the 3D urban macro scenario (3D-UMa) where antennas are positioned higher than the average height of buildings This allows more directed transmissions to line of sight (LOS) UEs in buildings or non-line of sight (NLOS) UEs obstructed by a building Next scenario is the 3D urban micro scenario (3D-UMi) where the 2D antenna array is assumed to be m from the ground with building height reaching higher than antenna height Due to the height of buildings, an enb in 3D-UMi case has a relatively wider elevation angle It has been shown in recent technical report 3GPP that 3D propagation effects of a wireless channel is modeled for both a large-scale and a small-scale fading [9] For large-scale parameters, pathloss model and LOS probability are defined as a function of UE height For small-scale fading model, elevation departure and arrival angle are modeled as a function of the UE height and the distance from the enb Based on measurement, the distribution of elevation angular spread is limited to -5 degree This reduces inter-user interference between two vertical beams and increases the possibility of multi user transmission in vertical direction C Performance Results ) Downlink Results: In Fig 4, 5, and 6, we plot the downlink performance results for deployment scenarios We clearly deserve that the cell throughput of 2D transceiver architecture is higher than that of the conventional horizontal beamforming Further, 2D 8Tx transceiver has marginal performance difference to D 8Tx transceiver in 3D-UMa 2m and 3D-UMi 2m scenario From enb scheduling perspective, the desired vertical angle would be large with smaller inter-site distance (ISD) Also, the scheme 35 3 25 2 5 (V,2H) (V,4H) (V,8H) Avg Tput (3D-UMa 5m) 5% edge Tput (3D-UMa 5m) 6 4 3 2 Fig 4 Downlink evaluation results for 3D-UMa with 5m ISD 25 7 2 6 5 4 3 2 Avg Tput (3D-UMa 2m) 5% edge Tput (3D-UMa 2m) (V,2H) (V,4H) (V,8H) Fig 3 Vertical rank beams for scheme and 2 Fig 5 Downlink evaluationresults for 3D-UMa with 2m ISD 2347

2 provides more 3D beamforming gain in terms of 5% edge performance As shown in Fig 3, from zenith ( ) to ground (8 ), the scheme 2 can use wide range of vertical angle and its resolution can be sharper than that of scheme With, only four precoding weights can be applied in vertical direction with 2Tx LTE codebook In case of 3D- UMi deployment, both scheme and 2 perform slightly better than D 8Tx This gain is mainly from the multi-user scheduling in vertical domain where enb antenna is placed lower than UE location As shown in the results, when the cell deployment size is smaller, more 3D beamforming gain can be achieved 2) Uplink Results: Fig 7 summarizes uplink average cell throughput and 5%-tile edge UE throughput for various antenna configurations and deployment scenarios; 2 and 7 for average and 5%-tile UE throughput with (4V, 2H) From the results, we can see that 2D transceiver architecture can enhance both average cell throughput and edge UE throughput In Fig 7, the performance gains for edge UEs are even larger than those for center UEs In uplink, enb tries to keep a balance between the received SINRs of each UE at a given rate through open-loop power control parameters In contrast to downlink case, available bandwidth for edge UEs is restricted in uplink so that the uplink edge 3 25 2 5 Fig 6 Fig 7 35 3 25 2 5 (V,2H) (V,4H) (V,8H) Avg Tput (3D-UMi 2m) 5% edge Tput (3D-UMi 2m) Downlink evaluation results for 3D-UMi with 2m ISD (V,2H) Avg Tput (3D-UMa 5m) Avg Tput (3D-UMa 2m) Avg Tput (3D-UMi 2m) 5% edge Tput (3D-UMa 5m) 5% edge Tput (3D-UMa 2m) 5% edge Tput (3D-UMi 2m) (V,4H) (V,8H) (4V,2H) (2V,4H) Uplink evaluation results for 3D-UMa and 3D-UMi scenarios 7 6 4 3 2 4 35 3 25 2 5 5 TABLE I Parameter SIMULATION PARAMETERS AND ASSUMPTIONS Value Multi-cell layout 9 enb each with 3 cells in hexagon layout Inter-site distance 2 m, 5 m for 3D UMa and 2 m for 3D UMi Tx power 43 dbm (3D UMa), 4dBm (3D UMi) Carrier frequency 2 GHz with MHz Number of UEs per cell UEs per cell UE antennal 2 Rx for downlink, Tx for uplink HARQ scheme IR asynchronous retransmission Link adaptation LTE MCS selection with % initial BLER CSI feedback Subband CQI with wideband RI and subband PMI Channel estimation Non-ideal for DL and Ideal for UL Antenna element pattern 65-3dB beamwidth in both vertical and horizontal Uplink power control Open-loop control with P =-9dB and α = 8 UE throughput does not increase even if more frequency resources are allocated However, it is possible to significantly enhance edge UE performances by adopting vertical dynamic beamforming Note that 5% edge UE throughputs of and s are higher than that of legacy (V, 8H) although the number of antenna elements are the same as the legacy MIMO configuration V CONCLUSIONS This paper discusses 3D beamforming operation with 2D transceiver architecture in 8Tx LTE-A, in terms of characteristics, deployment scenarios, and CSI feedback Furthermore, this paper studied the performance gain of 3D beamforming with the proposed schemes under 3D spatial channel properties Simulation results indicate that 2D 8Tx improves the user throughput over D s In addition, 2D 8Tx can provide marginal performance difference over the conventional 8Tx s while keeping same antenna dimension in 2 or 4 Tx s REFERENCES [] Cisco, Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 22-27, 23 [2] C Lim, T Yoo, B Clerckx, B Lee and B Shim, Recent trends in MU-MIMO," IEEE Commun Mag, vol 5, no 3, pp 27-35, Mar 23 [3] Y Kim, H Ji, J Lee, Y H Nam, B L Ng, I Tzanidis, Y Li and J Zhang, Full dimension MIMO (FD-MIMO): The next evolution of MIMO in LTE s," IEEE Wireless Commun Mag, vol 2, no 3, pp 92-, Jun 24 [4] F Athley, M Alm, O Kaspersson, K Werner, J Furuskog and B Hagerman, Dual-polarized base station antenna configurations for LTE," Proc IEEE APUSURSI, Sep 2 [5] W Xie, T Yang, X Zhu, F Yang and Q Bi, Measurement-based evaluation of vertical separation MIMO antennas for base station," IEEE Antenna and Wireless Propag Lett, vol, pp 45-48, Apr 22 [6] 3GPP RAN4#7, R4-4346, Radio Frequency (RF) requirement background for Active Antenna System (AAS) Base Station (BS)," May 24 [7] E Dahlman, S Parkvall and J Skold, 4G: LTE/LTE-Advanced for Mobile Broadband, Academic Press, 2 [8] Y Jiang, M K Varanasi and J Li, Performance analysis of ZF and MMSE equalizers for MIMO s: An in-depth study of the high SNR regime", IEEE Trans Inf Theory, vol 57, no 4, pp 28-226, Apr 2 [9] 3GPP Technical Reports 36873, Study on 3D channel model for LTE", Feb 24 2348